“Future House has released a set of scientific benchmarks called LabBench to measure the ability of AI agents to perform various scientific tasks.”